Automatically resolving textual questions that are numerically-based presents multiple challenges. For example, existing approaches cannot detect variables nor automatically normalize relevant units. Accordingly, improved techniques are needed.